whisper-speech-small

This model is a fine-tuned version of openai/whisper-small on the Google speech dataset. It achieves the following results on the evaluation set:

• Loss: 0.0021
• Wer: 0.3861

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 250
• training_steps: 3000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0095	0.2	250	0.0042	0.7239
0.0068	0.4	500	0.0051	1.0135
0.0045	0.6	750	0.0021	0.3861
0.0056	0.8	1000	0.0018	1.5927
0.0021	1.0	1250	0.0023	8.8803
0.0081	1.2	1500	0.0033	2.0270
0.0056	1.4	1750	0.0023	6.1293
0.0028	1.6	2000	0.0017	0.8687
0.0064	1.81	2250	0.0011	0.8687
0.0005	2.01	2500	0.0014	2.0270
0.0015	2.21	2750	0.0013	1.4961
0.0012	2.41	3000	0.0014	1.8822

Framework versions

• Transformers 4.30.0.dev0
• Pytorch 2.0.1+cu118
• Datasets 2.12.0
• Tokenizers 0.13.3